[Gluster-users] mtime handling seems generally buggy for directories

[Stephan von Krawczynski]

On Tue, 28 Jul 2009 05:34:55 -0500 (CDT)
Vikas Gorur <vikas at gluster.com> wrote:

> 
> ----- "Stephan von Krawczynski" <skraw at ithnet.com> wrote:
> 
> > Simply try this:
> > untar some kernel archive on your local disk and look at the mtime of
> > the created top directory. now untar the same archive on an exported gluster fs
> > and compare the mtimes.
> 
> > You will find out that mtime on gluster fs is generally not set (by
> > tar), not only during a healing process. Obviously this is a bug, easy to
> > reproduce.
> 
> I tried it just now and mtime seems to be set properly for me.
> 
> Can you run both server and client in TRACE log level and reproduce this?
> Also please send us both client and server log files and an strace dump
> of the tar extraction (you might want to pick a very small archive so that
> the strace isn't too big).
> 
> (Sorry for previous mail - hit send instead of save to drafts).
> 
> Vikas
> -- 
> Engineer - http://gluster.com/

Please investigate this example again.
1) Look at the server side fs. What mtimes do you see?
2) open a _new_ shell (xterm or else) on the extracting client. Does the mtime
look different there?

We see mtimes on servers always following the creation or last write. It is
never set by whatever packer we try (not just tar). Files are ok, dirs not.

Can this be some problem with fuse? We use the version delivered with opensuse
11.1 (2.7.2).

-- 
Regards,
Stephan